When assessing an AI visibility offering, do the same thing you would when evaluating any business tracking service - before signing, ask and consider these eight factors. Then give extra emphasis to the two factors you can check on paper (1) measurement methodology (what questions, what surfaces, how many runs, how often, what’s in the report) and (2) commercial terms (what’s exactly guaranteed, for which surfaces, what if you walk away). All the “outcome” claims about what you’ll get (“we’ll get you into the answers”, AI traffic projected, client results mentioned) are impossible to verify against what’s published by any engine - so take it as opinion that you should discount. The range is from self-serve trackers ($79/month-ish) to done-for-you local agencies ($999+/month) and the two anchor examples listed here checked out vs vendors’ own pages on 2026-11-04. $79-$249/month Lighthouse Local and $999/month KKV Labs … this post isn’t about choosing between them. It’s about giving you the questions that you can apply to any of the offers out there.
The market in two sentences
For local businesses there are two versions of the offer. The self-service trackers that ask you questions at regular intervals, record the responses of each engine, provide a scorecard and to-do list, and leave the fixes up to you (normally $79-$249/month, first try it, no obligation). Or done-for-you agencies that provide measurement along with the fixes (website changes, profile changes, citations, content, review management) for $999+/month (vendor-quoted prices 2026-11-04), often a set of questions that’s fixed, plus a minimum term, plus a promise to perform. The two versions are answering the same measurement problem from how to measure ai visibility for local business. For whether or not you should buy this channel in the first place (i.e. the demand side of the problem) see do customers find local businesses on chatgpt (a few names in a shortlist, significant run-to-run variation, and no official dashboard, so the verification takes place in the methodology part of the offer not the outcomes part).
The eight checks
1. Which surfaces, and does the offer say so specifically?
“How many AI platforms?” is a checklist. “Which ones? Do I care which ones?” Different platforms - the search stacks are their own (ChatGPT, Perplexity, Claude) for some, some (Gemini, google ai features) have Google’s index, different location logic (engine breakdown). If a deal is “just one or two surfaces”, it’s a partial number, and okay if they say it. But if they say “all ai” without naming the engines, no.
2. Is the question set fixed, and agreed in writing?
Only if the questions are locked across months. How many questions? Who picks them? Can you see/approve them ahead of time before you start working? If you want to change a question during a term. Etc. If you don’t have a written agreed question set, then there’s no sane way to give any guarantee (a guarantee that isn’t tied to specific questions is a guarantee that isn’t tied to anything).
3. How many runs, on what cadence, and how is variance handled?
This is the key difference between measuring vs taking screenshots. The variation in repeated exactly the same query - between 20-33% (BrightLocal Local AI Visibility Study (September 2026. - vendor study. brightlocal.com/research/local-ai-visibility-study/). If you only run 1 for the same question each month, you aren’t learning much. What’s the frequency they run for each question. What cadence (daily check? Weekly sweep? bi-weekly re-runs, depends on vendor name of cadence)? What’s a real change vs a noise? If they give you median/rate as answer (“4 of 6 times named”). They are using the right unit. If they send you a screenshot they aren’t.
4. What does the report actually contain?
Sample report should be shown before you pay. Actual or clearly marked as a sample. So… we’re looking for verbatim answer text or exact strings that were mentioned. What each engine said about you (mistakes - wrong hours, out of date prices, wrong services stated). What surfaces they mentioned. What competition you were compared with. Month over month deltas. Screenshots only - demo. With captured text - scorecard - report. Rules for what should and should not be in the monthly report with those numbers, are here in report template post.
5. Pricing model: all-inclusive, tiered, or add-on-driven?
Three “honestly” approaches – flat fee for everything (the KKV thing of “everything for $999 a month. nothing ala carte”), published-limits tier plans (Lighthouse – e.g., here are the published limits on the number of prompt/site/competitors/seats in each plan…with add-ons priced above it), and base price + project-specific services quoted. Any of these can be honest. The key is whether they published the limits before you’re signing. (Yes, it’s fine if they have a fixed 5 question set, or a prompt-cap for weekly sweep, or a limit on the rebuild of six pages. So long as that’s in writing. Because otherwise, you’re going to know in month 3.)
6. The guarantee: what exactly, on which surfaces, with which exit ramp?
G’s are contracts, interesting bits are clauses. In the case of KKV Labs, they have really specific published terms. Like: 3 out of 5 agreed questions recommended by day 90 (ChatGPT + Google AI Mode). Or get your money back. And they also tell you in 24 hours from the baseline if they decide the market is unwinnable for you and you don’t pay (vendor published terms, kkvlabs.com (accessed 2026-11-04)). So. Look at any similar G. You’ll see: exact bar (what’s recommended, name? cite? position?). Surfaces they apply to. Measuring method (and this needs to be the same runs they report back to you). What if the vendor itself says the market is unwinnable.
7. Exit terms: what happens to the work and the data?
Two different questions: Work: If you have changes to site/profile/content, is what you got still yours when you leave at 31 days of the month? Per KKV, “stays yours, even if you leave/take a refund” (vendor-published, last access 2026-11-04). You can try to find that phrase in any competition terms. Otherwise you get a site that works only as long as you pay.
Data: Are you able to get your history/prompt pool/scores out? If you have a tracker that its value is 18 months data, you should have it after canceling it with data.
8. Claims that should not survive a 30-second check
The following 3 points don’t make it past the engines’ documentation (it shouldn’t take a bibliophile to smell the bullshit):
- “We submit your business to ChatGPT”, etc. No engine documents the submission/opt-in for local businesses. Google’s even explicit: there’s nothing more you need to do to show up in their AI stuff (docs 2026-11-04). Here’s the ranked list of fixes – see what’s on it (and not on it): how to get your local business recommended by AI
- AI traffic/leads projections No one’s actually published the AI answer click volume for local businesses. Any number you hear is a projection, not something we’ve measured (the honest state of AI referral data)
- Using un-sourced stats It’s normal for vendors to show off third-party research – like on KKV’s homepage, which mentions BrightLocal, Semrush, Bain, Adobe Analytics. But make sure you look at who/when/vendor-commissioned the original study (so here’s the deets on the BrightLocal stuff: brightlocal sell a local visibility platform. their 45% usage among consumers comes from a survey they did in feb 2026 of 1002 us adults, brightlocal.com/research/lcrs-ai-trust/). So if their own research is what’s supposed to sway you, make sure you ask them about their methodology before assuming that number is data.
Two anchors, fact-checked
Both of these pages were viewed on 2026-11-04. All vendor claims are actual vendor claims – not our quality opinion.
Lighthouse Local - self-serve tracker
$79/mth (Starter), $249/mth (Growth). Monthly payments. Custom enterprise $599+? Free trial 14-days. No commitment. Cancel at any time. lighthouselocal.ai/pricing (2026-11-04). Track all 5 (ChatGPT, Claude, Gemini, Perplexity, Google AI) for all prompts. Starter: 50 tracked prompts/weekly sweep + 3/day. 3 competitors. 30-days history. Growth: 150 tracked prompts/weekly sweep + 12/day. 9 competitors. 90-days history. Not metered – just hard limits. GBP management / citations / on-page / links – as done-for-you add-ons at additional cost. Agency/white-label. Check before you pay for it: how many times they actually score a given prompt in a sweep (variance, 3)? What do they mean by “plain english action plan”? (PDF? Tickets? Specialist?) Do you get to export your history and prompt pool on cancel? (7)
KKV Labs, done for you agency. $999/month. no add-ons, everything included. 24 hr baseline within. 5 questions written down agreed at kickoff. 90 day build. ai-readable site fixes or rebuild up to 6 pg. LocalBusiness / Service / FAQ structured data. gbp + bing places refresh. category dir listings. 2-4 new answer pages per month. review request engine. biweekly re-run same 5 q on chat gpt and google ai mode. screenshots, scorecard. terms: 3 mo min then mo-to-mo. recommened 3 of 5 by 90 days or full refund. guarantee. (we can’t win the market, here you go, you don’t pay anything.) 1 business/indus/city. all built is yours. ( kkvlabs.com – accessed 2026-11-04 ) Before paying make sure to check: written 2-4 pages per month. rebuild scope. how are 5 questions selected? can you replace one in-term? what exactly is the “trusted” stage. (local press? reddit? mention on review sites?) these are 3rd party surfaces, and their approach to that is what you can’t see from no docs. what is the ‘recommended’ in scorecard for day 90? (named? cited? position?)
You can’t independently verify the numbers from case studies (of either vendor) from the pages, including kkv which the vendor itself says are their properties and not external clients. And that’s the whole point. The anchors are helpful because you can check their terms, not their outcomes.
The five questions to ask any offer
- show me what a sample report looks like. what is being captured per answer exactly. including what the engines said wrong?
- how often are you running each question and on what schedule. how are you determining that this is real change and not run to run variance?
- which surfaces are guaranteed. what’s the definition of “recommended” in your scorecard?
- if we roll out of there in month 4, what happens to the work and the data?
- which engine documentation is your core service based on? which documentation is your “submission” service based on, since none of them have a submission path published? If they can’t write up 2 and 3 as part of the offer, you’re buying a screenshot. If you’d like to benchmark any offer against your own numbers before you commit to either vendor shape, the lowest cost way to do that is with the self-run baseline test. Just an hour, 10-20 questions, 4 engines, and a before state that you control. Or if you’d like that baseline run, logged, and then re-run on a monthly basis with a named competitor set and the whole process documented before any changes are made, see the methodology. If you’re on the agency side of this market and not a single business, you’ll find the white-labeled version of this very same measurement/reporting loop in white label ai visibility ops